This is a public adaptation of a letter I wrote to everyone at EnergyQuant in August 2026. It also develops ideas from my short note first published that June.
Understanding a problem is part of solving it.
AI makes it easier to retrieve information, compare approaches, write code, and analyse data. As the first answer arrives faster, I care more about whether we have understood the real problem, asked the right questions, judged the answer's reliability, and turned intelligence into a useful result.
Experience can help our judgment, and it can also constrain it. When a familiar method arrives before we have understood the problem, we can force a new situation into an old frame. I remind myself to understand the new problem before letting an old method decide for it.
Be clear about the problem
Before starting a task, explain whose problem it addresses, why it needs solving, and what would count as completion.
The actual need, business setting, system boundaries, constraints, and acceptance criteria belong in the same discussion. Clear conditions let us put technology to useful work. If we start before understanding the problem, AI may help us solve the wrong problem faster.
Defining a problem has value in its own right. It tells us which information matters, which approaches deserve comparison, and how we can check the result.
Investigate and reuse before building what is missing
Most general needs have appeared before. Existing documents, code, past projects, mature products, open-source tools, and research often contain part of the answer.
Once the need is clear, look at the capabilities already available inside and outside the organisation. Then decide what can be reused, what needs adapting, and what genuinely needs building.
Investigation helps us avoid repeating work at a low level. It reveals the smallest important gap between what already exists and what the task actually requires. Understanding existing work is itself a form of learning.
Experience needs this check too. Knowing a method well does not make it suitable for every object, decision time, or constraint.
Carry the work through to a useful result
AI lowers the barrier to ordinary work and raises the standard for excellent work. After producing a document, a piece of code, or a page quickly, we still need to ask whether it is reliable, solves the original problem, and can actually be used.
Pursuing excellence does not mean getting everything right on the first attempt or simply working longer hours. It means being dissatisfied with unclear problems, duplicated work done without investigation, and results that are only approximately usable.
Exploration and trial and error are necessary. Each exploration should add understanding. Each solved problem should, where possible, leave a reusable capability and a better starting point for the next task.
Understand the problem. Investigate and reuse before building what is genuinely missing. Turn intelligence into results that can be checked.